Biology subjects

Suphavilai, C.

Publications and source records attributed to Suphavilai, C..

7 recordsLinked to original sources

bioRxiv2024

Direct high-throughput deconvolution of unnatural bases via nanopore sequencing and bootstrapped learning

The discovery of non-canonical bases (NCBs) in viruses and the development of synthetic xeno-nucleic acids (XNAs) to expand the genetic alphabet has spawned interest in many applications, from viral genomics, to synthetic biology and DNA storage. However, the inability to read non-canonical bases in a direct, high-throughput manner has been a significant limitation to its study and applicability. Here we demonstrate that XNA templates containing non-canonical bases can be directly and robustly sequenced (>2.3 million reads/flowcell, similar to DNA controls) on a MinION sequencer from Oxford Nanopore Technologies to obtain signal data that is significantly distinct from DNA controls (median fold-change >6x). To enable training of machine learning models that deconvolve these signals and basecall non-canonical and canonical bases, we developed a framework to synthesize a complex pool of 1,024 NCB-containing oligonucleotides with diverse 6-mer sequence contexts and high purity (>90% NCB-insertion on average). Bootstrapped models to assist in data preparation, and data augmentation with spliced reads to provide high context diversity, enabled learning of a generalizable model to call canonical as well as non-canonical bases with high accuracy (>80%) and specificity (99%). These results highlight the versatility of nanopore sequencing as a platform for interrogating nucleic acids for viral genomic and xenobiology applications, and the potential to transform the study of genetic material beyond those that use canonical bases.

synthetic biology↗

bioRxiv2023

MetageNN: a memory-efficient neural network taxonomic classifier robust to sequencing errors and missing genomes

BackgroundWith the rapid increase in throughput of long-read sequencing technologies, recent studies have explored their potential for taxonomic classification by using alignment-based approaches to reduce the impact of higher sequencing error rates. While alignment-based methods are generally slower, k-mer-based taxonomic classifiers can overcome this limitation, potentially at the expense of lower sensitivity for strains and species that are not in the database. ResultsWe present MetageNN, a memory-efficient long-read taxonomic classifier that is robust to sequencing errors and missing genomes. MetageNN is a neural network model that uses short k-mer profiles of sequences to reduce the impact of distribution shifts on error-prone long reads. Benchmarking MetageNN against other machine learning approaches for taxonomic classification (GeNet) showed substantial improvements with long-read data (20% improvement in F1 score). By utilizing nanopore sequencing data, MetageNN exhibits improved sensitivity in situations where the reference database is incomplete. It surpasses the alignment-based MetaMaps and MEGAN-LR, as well as the k-mer-based Kraken2 tools, with improvements of 100%, 36%, and 23% respectively at the read-level analysis. Notably, at the community level, MetageNN consistently demonstrated higher sensitivities than the previously mentioned tools. Furthermore, MetageNN requires < 1/4th of the database storage used by Kraken2, MEGAN-LR and MMseqs2 and is >7x faster than MetaMaps and GeNet and >2x faster than MEGAN-LR and MMseqs2. ConclusionThis proof of concept work demonstrates the utility of machine-learning-based methods for taxonomic classification using long reads. MetageNN can be used on sequences not classified by conventional methods and offers an alternative approach for memory-efficient classifiers that can be optimized further.

bioinformatics↗

bioRxiv2021

Network-guided supervised learning on gene expression using a graph convolutional neural network

BackgroundTranscriptomic profiles have become crucial information in understanding diseases and improving treatments. While dysregulated gene sets are identified via pathway analysis, various machine learning models have been proposed for predicting phenotypes such as disease type and drug response based on gene expression patterns. However, these models still lack interpretability, as well as the ability to integrate prior knowledge from a protein-protein interaction network. ResultsWe propose Grandline, a graph convolutional neural network that can integrate gene expression data and structure of the protein interaction network to predict a specific phenotype. Transforming the interaction network into a spectral domain enables convolution of neighbouring genes and pinpointing high-impact subnetworks, which allow better interpretability of deep learning models. Grandline achieves high phenotype prediction accuracy (67-85% in 8 use cases), comparable to state-of-the-art machine learning models while requiring a smaller number of parameters, allowing it to learn complex but interpretable gene expression patterns from biological datasets. ConclusionTo improve the interpretability of phenotype prediction based on gene expression patterns, we developed Grandline using graph convolutional neural network technique to integrate protein interaction information. We focus on improving the ability to learn nonlinear relationships between gene expression patterns and a given phenotype and incorporation of prior knowledge, which are the main challenges of machine learning models for biological datasets. The graph convolution allows us to aggregate information from relevant genes and reduces the number of trainable parameters, facilitating model training for a small-sized biological dataset.

bioinformatics↗

bioRxiv2020

TUGDA: Task uncertainty guided domain adaptation for robust generalization of cancer drug response prediction from in vitro to in vivo settings

Over the last decade, large-scale cancer omics studies have highlighted the diversity of patient molecular profiles and the importance of leveraging this information to deliver the right drug to the right patient at the right time. Key challenges in learning predictive models for this include the high-dimensionality of omics data, limitations in the number of data points available, and heterogeneity in biological and clinical factors affecting patient response. The use of multi-task learning (MTL) techniques has been widely explored to address dataset limitations for in vitro drug response models, while domain adaptation (DA) has been employed to extend them to predict in vivo response. In both of these transfer learning settings, noisy data for some tasks (or domains) can substantially reduce the performance for others compared to single-task (domain) learners, i.e. lead to negative transfer (NT). We describe a novel multi-task unsupervised domain adaptation method (TUGDA) that addresses these limitations in a unified framework by quantifying uncertainty in predictors and weighting their influence on shared domain/task feature representations. TUGDAs ability to rely more on predictors with low-uncertainty allowed it to notably reduce cases of negative transfer for in vitro models (63% for drugs with limited data and 94% overall) compared to state-of-the-art methods. For domain adaptation to in vivo settings, TUGDA improved performance for 6 out of 12 drugs in patient-derived xenografts, and 7 out of 22 drugs in TCGA patient datasets, despite being trained in an unsupervised fashion. TUGDAs ability to avoid negative transfer thus provides a key capability as we try to integrate diverse drug-response datasets to build consistent predictive models with in vivo utility. Availabilityhttps://github.com/CSB5/TUGDA

bioinformatics↗

bioRxiv2020

Predicting heterogeneity in clone-specific therapeutic vulnerabilities using single-cell transcriptomic signatures

While understanding heterogeneity in molecular signatures across patients underpins precision oncology, there is increasing appreciation for taking intra-tumor heterogeneity into account. Single-cell RNA-seq (scRNA-seq) technologies have facilitated investigations into the role of intra-tumor transcriptomic heterogeneity (ITTH) in tumor biology and evolution, but their application to in silico models of drug response has not been explored. Based on large-scale analysis of cancer omics datasets, we highlight the utility of ITTH for predicting clinical outcomes. We then show that heterogeneous gene expression signatures obtained from scRNA-seq data can be accurately analyzed (80%) in a recommender system framework (CaDRReS-Sc) for in silico drug response prediction. Patient-derived cell lines capturing transcriptomic heterogeneity from primary and metastatic tumors were used as in vitro proxies for validating monotherapy predictions (Pearson r>0.6), as well as optimal drug combinations to target different subclonal populations (>10% improvement). Applying CaDRReS-Sc to the increasing number of publicly available tumor scRNA-seq datasets can serve as an in silico screen for further in vitro and in vivo drug repurposing studies. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=191 SRC="FIGDIR/small/389676v1_ufig1.gif" ALT="Figure 1"> View larger version (32K): org.highwire.dtl.DTLVardef@1aa5795org.highwire.dtl.DTLVardef@5c2323org.highwire.dtl.DTLVardef@106e074org.highwire.dtl.DTLVardef@7a0834_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LILarge-scale analysis to establish the impact of transcriptomic heterogeneity within tumors on clinical outcomes C_LIO_LICalibrated recommender system for drug response prediction based on single-cell RNA-seq data (CaDRReS-Sc) C_LIO_LIPrediction of drug response in patient-derived cell lines with transcriptomic heterogeneity C_LIO_LIIn silico identification of drug combinations that work based on clonal vulnerabilities C_LI

bioinformatics↗

bioRxiv2017

Predicting Cancer Drug Response Using a Recommender System

MotivationAs we move towards an era of precision medicine, the ability to predict patient-specific drug responses in cancer based on molecular information such as gene expression data represents both an opportunity and a challenge. In particular, methods are needed that can accommodate the high-dimensionality of data to learn interpretable models capturing drug response mechanisms, as well as providing robust predictions across datasets.\n\nResultsWe propose a method based on ideas from \"recommender systems\" (CaDRReS) that predicts cancer drug responses for unseen cell-lines/patients based on learning projections for drugs and cell-lines into a latent \"pharmacogenomic\" space. Comparisons with other proposed approaches for this problem based on large public datasets (CCLE, GDSC) shows that CaDRReS provides consistently good models and robust predictions even across unseen patient-derived cell-line datasets. Analysis of the pharmacogenomic spaces inferred by CaDRReS also suggests that they can be used to understand drug mechanisms, identify cellular subtypes, and further characterize drug-pathway associations.\n\nAvailabilitySource code and datasets are available at https://github.com/CSB5/CaDRReS\n\nContactnagarajann@gis.a-star.edu.sg\n\nSupplementary informationSupplementary data are available online.

bioinformatics↗

bioRxiv2017

ConsensusDriver Improves Upon Individual Algorithms For Predicting Driver Alterations In Different Cancer Types And Individual Patients -- A Toolbox For Precision Oncology

BackgroundIn recent years, several large-scale cancer genomics studies have helped generate detailed molecular profiling datasets for many cancer types and thousands of patients. These datasets provide a unique resource for studying cancer driver prediction methods and their utility for precision oncology, both to predict driver genetic alterations in patient subgroups (e.g. defined by histology or clinical phenotype) or even individual patients.\n\nMethodsWe performed the most comprehensive assessment to date of 18 driver gene prediction methods, on more than 3,400 tumour samples, from 15 cancer types, to determine their suitability in guiding precision medicine efforts. These methods have diverse approaches, which can be classified into five categories: functional impact on proteins in general (FI) or specific to cancer (FIC), cohort-based analysis for recurrent mutations (CBA), mutations with expression correlation (MEC) and methods that use gene interaction network-based analysis (INA).\n\nResultsThe performance of driver prediction methods varies considerably, with concordance with a gold-standard varying from 9% to 68%. FI methods show relatively poor performance (concordance <22%) while CBA methods provide conservative results, but require large sample sizes for high sensitivity. INA methods, through the integration of genomic and transcriptomic data, and FIC methods, by training cancer-specific models, provide the best trade-off between sensitivity and specificity. As the methods were found to predict different subsets of drivers, we propose a novel consensus-based approach, ConsensusDriver, which significantly improves the quality of predictions (20% increase in sensitivity). This tool can be applied to predict driver alterations in patient subgroups (e.g. defined by histology or clinical phenotype) or even individual patients.\n\nConclusionExisting cancer driver prediction methods are based on very different assumptions and each of them can only detect a particular subset of driver events. Consensus-based methods, like ConsensusDriver, are thus a promising approach to harness the strengths of different driver prediction paradigms.

bioinformatics↗